Failure modes for autonomous vehicles are associated with different levels of risk that the failure mode will result in an undesirable outcome such as a collision or vehicle breakdown. Some failure modes require that the vehicle immediately pull off of the road in order to minimize the risk. In other cases, a failure mode does not impose an immediate risk, and the vehicle can continue. Analyzing the level of risk associated with a failure mode, identifying alternatives for managing the failure mode, and evaluating risk associated with each of the alternatives, require a substantial amount of computing power and data. Additionally, some data, such as weather data, traffic data, and availability of resources, is changing on a continuous basis. It may not be practical to supply the computing power, or to supply the data in a timely manner, to each vehicle that needs to manage a failure mode.